Style Manipulation Attack on Qwen3-8B
2ScoreJudgeDeceiver
Evaluation Results
| Method | Links | |
|---|---|---|
| JudgeDeceiverOptimization Steps=600, Attack Type=White-box2026.05 | 2 | |
| BITEOptimization Steps=25, Attack Type=Black-box2026.05 | 1.905 | |
| JudgeDeceiverOptimization Steps=300, Attack Type=White-box2026.05 | 1.588 | |
| JudgeDeceiverOptimization Steps=200, Attack Type=White-box2026.05 | 1.438 | |
| JudgeDeceiverOptimization Steps=100, Attack Type=White-box2026.05 | 0.889 | |
| JudgeDeceiverOptimization Steps=25, Attack Type=White-box2026.05 | 0 |